基于边缘盒与低秩背景的图像显著区域检测算法  

Research on Image Salient Regions Detection Combing Edge Boxes and Low-rank Background

作　　者：申瑞杰[1] 张军朝[2] 郝敬滨 SHEN Rui-jie;ZHANG Jun-chao;HAO Jing-bin(School of Computer,Jiangsu Normal University,Xuzhou 221003,China;College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030024,China;Schoo1 of Computer Science and Technology,China University of Mining and Technology,Xuzhou 221003,China)

机构地区：[1]江苏师范大学计算机学院,徐州221003 [2]太原理工大学电气与动力工程学院,太原030024 [3]中国矿业大学计算机科学与技术学院,徐州221003

出　　处：《计算机科学》2017年第12期298-303,共6页Computer Science

基　　金：江苏省教育科学"十二五"规划课题(C-c/2011/02/010);江苏省教育科学"十二五"规划2013年度立项课题(D/2013/02/273)的阶段性成果;山西省重大专项项目(20131101029)资助

摘　　要：针对现有显著性区域边界不明确和检测效果鲁棒性较差等问题,提出了一种新颖的图像显著区域检测方法,该方法结合了边缘盒粗定位和低秩背景模型细筛选来提高显著区域的检测性能。首先,对基于边缘盒的图像显著区域检测方法进行改进,采用OTSU方法自适应计算边缘模值的最佳分割阈值,以替代固定分割阈值,降低边界点检测误差;其次,在基于边缘盒检测到的可疑显著区域上,采用鲁棒主成分分析方法获取图像的低秩分量,构建背景模型,并基于背景差分方法剔除背景区域,减少显著区域的虚检现象。在PASCAL VOC 2007数据集上的实验结果表明,提出的方法明显提高了显著区域检测的精确度和召回率,同时具有较高的检测效率。Aiming at the problem that traditional saliency detection methods suffer from unclear boundary and bad robust detection performance,an novel image salient regions detection method was proposed,combining two detection stages including edge boxes for rough location and low-rank background model for refining,to enhance the performance of salient regions detection.First,it improves the image salient regions detection method based on edge boxes.It uses OTSU method for adaptively computing the optimal threshold value of edge magnitude,to replace fixed threshold method and reduce boundary detection error.Second,on the basis of suspicious salient regions detected by edge boxes based method,it uses robust principal component analysis method to obtain the low-rank component of the image for building a background model,and eliminates the background regions based on background subtraction method to reduce false detection of salient regions.Experimental results on the PASCAL VOC2007dataset show that,this method can significantly improve the precision and recall metrics of salient regions detection,and has higher detection efficiency.

关 键 词：显著区域检测 边缘盒 鲁棒主成分分析 低秩背景 OTSU 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]